EVALUATION OF SOYBEAN PLANT SHAPE-BASED ON TREE-BASED MODELS

Plant shape, an important factor in soybean plant breeding, is current ly evaluated visually by soybean plant breeders, often making judgment unstable and subjective. The purpose of our study was to create proce dure for objectively evaluating soybean plant shape. Features of shape were determined by image analysis. Tree-based models based on recursi ve partitioning were then used to categorize shapes into three classes - ''good,'' ''fair'' and ''poor'' - or two classes - ''good'' and ''n ot good.'' Classification results based on tree-based models demonstra ted highly acceptable predictability. Although model-based performance attained approximately the same discriminatory level as conventional linear discriminant function, it had the distinct advantage of outstan ding interpretability, with shape parameters in the best predictive tr ee-based model with those selected empirically by expert breeders.